Documente Academic
Documente Profesional
Documente Cultură
I 185
CURRENT PAPERS
1971
.a C.P. No.1185
November 1969*
bY
P. R. Edwards
In this Report factors affecting the accuracy of Miner's Rule are dls-
cussed. An nvestlgation 1s also described of the cumulative damage behaviour
of DID 5014 alumlnlun alloy lug specuwns using random loading. It 1s con-
cluded that the devlatlons from Miner's Rule observed in the uxestlgation can
be ascribed manly to the actlon of residual stresses associated with yielding
at tllf point of fatzgue lnltlatlon. An attempt is made to quantify this
effect.
M
Symbols 5
1 INTRODUCTION 7
5 CUMULATIVEDAMAGEPROGRAMME 25
.
CONTENTS (Contd.)
Page
-2 -2
Conversions: 1 ksi = 1000 lb(f) in = 6.894 MNm = 0.685 Hb .
5
SYMBOLS
SYMBOLS (Contd.)
0, o1 , c2 stress
1 INTRODUCTION
Many cases in which the rule does not give a good prediction of the
fatigue life of aluminium alloys can be explained by the influence of residual
stresses acting at the point of fatigue initiation. Such stresses can be
generated by local yielding at stress concentrations under the highest loads
applied or may be already present due to manufacturing processes. In section 3
of this Report the influence of residual stresses on cumulative damage
behaviour 1s discussed. In Appendices A - F other reasons for inaccurate pre-
diction are also discussed, including variation of relative damage rate with
stress, cycles below the fatigue limit, effect on life of the particular load
at which a component fails, and the effects of fretting.
fatigue life under constant amplitude and two types of variable amplitude load-
ing each at three mean stress levels. Measurements were made of residual static
strength at different stages in the fatigue life under the various types of
loading action. By this means the relative rates of accumulation of fatigue .
*In this Report 'spectrum' refers to a spectrum of amplitudes and not of fre-
quencxs as in 'power spectrum'.
9
achieved life
to the ratlo for random tests, and for
predicted life using Miner's Rule
programmed tests where the number of programmes to failure 1s large. 2 i IS
not equal to this ratlo for step tests except when 1; = 1. (See Appendix G.)
It 1s apparent f~ov a study of the literature that M~ner's Rule can give
highly inaccurate estunates of fetlgue life under variable amplitude loading.
23 24
For ln5tLLPCe, 7alues of 1 ; varying from 0.01 to 8 have been reported.
Geometric residual stresses can expialn much of the varlatlon in 1 g , and are
dlscussed III this sectlon. [Ither factors are dlscussed III the appendlces,
*iXIEly variation of relative danr.ge rate with stress and associated changes in
fatigue mechanism (Appendu A), effect of the load at which a component falls
(Appendix B) effect of low level stress cycles (Appendix C), effect of strain
hardening (Appendix D), effect of fretting (Appendix E) and crack propagation
conslderatxxw (Appendx? F).]
other materials and, in particular to many components XI medium and low strength
steels are explained in section 3.3. The behavxour mentioned above of notched
alumln~um alloy specimens can be explained as follows, simplified to apply
strictly only to a unlaxial stress state at the notch root.
.
Consider a notched specuwn having a stress concentration factor Kt ,
and made of a material havxng a stress-strain curve as shown in Fig.1. The
. .
application of a net mean stress 0 will give rise to a stress = Kt om
m 9
at the notch root. Assume that an alternating stress ~a is applied of such
a value that (cm + aa) K is less than o . In this case the material at the
t Y
root of the notch will cycle between 02 = (am + ua) Kt and o3 = (om - ua) Kt.
The local mean stress oem will of course be o1 . If oa 1s larger so that
(0 m + oa) K
t
> CI , then the ~~XUJUJI stress at the root of the notch under cyclic
Y
loading ~111 be given by 04 = (am + ua) Kp , where Kp 1s the stress concentra-
tion factor under conditions of yielding at the notch root. This is not a
sunple value to calculate since K is itself a function of the plastic strain.
P
However, an approximate value for 04 can be obtained by assuming that ~4 is
equal to the elastic strain, i.e.
(0 + aa)
E4
=Kt mE (2)
After c~4 is reached (on the fust quarter cycle of the net alternating
stress amplitude ua) the material at the root of the notch will cycle purely
elastically between o4 and where
O5
o5 = o4 - 2K; oa
where is the new elastic stress concentration factor for the notch
";
after the geometric distortion which accompanied the initial plastic strain.
For maximum strain values at the notch root of 2X, the error in assuming
Kt = Kt ' is negligible. Therefore
11
o5 = o4 - 2Kt oa
o6 = o4 - Kt oa . (3)
If the alternating stress 1s made larger the maximum local stress ~111
increase to say 0, , but the local mean stress 08 will be larger than
" - Kt o a due to local compressive yleldlng between A and B* . Cycling
7
at this level ~~11 therefore uwolve plastic yielding on every cycle unless the
hysteresis loop ABC 1s ultxnately closed by a cyclic strain hardenrng process.
Flg.2 shows the calculated variatwn in local mean stress with alternating
stress level under constant amplitude loading for the lug specunens used ln the
lnvestlgatlon reported in section 5 of this Report. It has been assumed that
the stress concentration acts as a pure strain concentration, 1.e. equation (2)
holds, and no allowance has been made for the effects of compressive yielding.'
Flg.3 shows the varlatlon in njm (Ref.32) measured on a large notched specunen
of ~65 with a stress concentration factor of 2.27 and demonstrates how local
compressive yleldxng can affect the resu:ts. The stress hutory at the notch
root was determined by a method used in Ref.33.
ohr = O4 - Op Kt (4)
where is now the maximum local stress corresponding to the peak stress
O4 OP.
Equatlon (4) ~111 be modified if local compressive yleldlng occurs giving a
larger value.of GLm as in Flg.3 for values of ~a greater than 8.8 ksl rms.
Under random loading, therefore, due to the presence of a compressive residual
XAfter initial plastic tensile yielding, local compressive yielding ~111 occur
at stresses much more 'tensile' than the initial compressive yield point. Simi-
larly, the yield point in tension is lowered by any prior plastic straining in
compression. This phenomenon 1s known as the Bauschlnger Effect (Ref.31).
12
stress, oEmr ~111 be lower than (or equal to) the local mean stresses which
existed under the constant amplitude loading actions used to obtain the part of
any S-N curve which were needed in turn to predict the life under random load-
ing. This results in less damage per cycle in the random loading case than is
predicted from the constant amplitude data. Hence Miner's Rule tends to be
conservative for notched specimens under symmetrical loading at positive mean
stress levels. At zero mean stress oem will be zero under both random and
sinusoidal loading so no correspondmg effect on 1 t is to be expected.
geometric residual stresses. These specimens are then taken as 'new' specmens
to be tested under variable and constant amplitude loading at a posltlve mean
stress to determzne the applicability of Miner's Rule. In the absence of local
compressive yielding, tensile local yielding will not occur under either type
of loading and so the value of 4 em will be the same for both constant ampli-
tude and variable amplitude loading actions. Hence 18 will be less than for
specimens which have not been preloaded because, UI the case of the preloaded
specimens, the beneficial effect of the residual stresses applies under constant
amplitude as well as under variable amplitude loading. An as illustration,
reduced values of 1 ; have been shown for preloaded lug spec~~ns under pro-
granrned loading 39 compared with values for specimens which were not preloaded.
After the pre-stressing, constant amplitude and random loading spectrum tests
were performed at different overall stress levels, the applied random spectrum
having an exponential distribution of load amplitudes. In Fig.4, the data
obtained has been replotted to show the variation in 1 i with alternating
stress at different pre-stress ratios and it can be seen that positive pre-
stress ratios giving compressive residual stresses result in low values of
1; and vzce versa.
In this section it has been shown that geometric residual stresses can be
of extreme importance in cumulative damage of aluminium alloy specimens, since
they can cause large variations in 1 E . All the examples quoted have xwolved
notched aluminium alloy specxr~ens since residual stresses can easily be induced
by local yielding at the notch root. However, it is clear that since machining
operations can induce residual stresses into the surface of plain specimens,
such variations in 1 a can apply to all classes of specimen.
14
For residual stresses to be effective m altering the fatigue life two con-
ditmns must be fulfilled. First, the stresses must be reasonably stable unde-
cyclic loading below the yield point. If such stresses are relaxed by low 21:5-i-
tude cycling, then local mean stresses in notched specimens "111 tend to red:ce
progressively under fatigue loading, regardless of the loading action. In this
case, any changes in the magnitude of %n which do occur, as described in the
last sectlon, due to local yielding may well be insignificant if the rate of
decay of oem 1s large enough. Secondly, the yield point of the material must
be considerably above Kt oa at the lowest value of oa which produces sigrii-
flcant fatigue damage oafa. . IfK t CJ af2 ' Oyp then the local mean stress in
a notched specimen will always be relatively close to zero under symmetrical
loading greater than the fatigue lxnit stress oafe , regardless of the value
of the net mean stress. In this case residual stresses cannot be sustalned
under cyclic loading of any slgniflcant amplitude because the first cycle after
the introduction of such stresses ~11 relax them by local yielding. The para-
meter governing whether or not this occurs 1s therefore the ratio
effect of the residual stresses on the fatigue life of the bars, demonstrating
a much greater effect for aluminium alloy than for mild steel as would be
expected from the stress measurements.
Many low and medium strength steel components are therefore not greatly
affected by residual stresses due to the low yield point of such steels com-
pared with the fatigue limit stress and the tendency for residual stresses to
be relaxed under low-level cyclic loadmg. This lack of sensitivity to
43-45
residual stresses has been further demonstrated in tests on medium
strength steel notched specimens, which have been shown to be affected rela-
tively little by mean stress or the application of single unidirectional high
loads. However as was stated in section 3.1 the arguments in section 3
16
It can be concluded then that geometric residual stresses may affect the
fatigue damage growth rate to some extent throughout the fatigue life, where
initiation 1s from a very large notch, i.e. where in the untracked specimen
the elastic stress value does not change significantly over the depth d in
Fig.8. For smaller notches of the same Kt value, cracks will propagate
through regions of significantly diminishing residual stress and hence such
stresses are likely to be most effective in the early stages of fatigue life.
For very small sized specimens where the stress field at the crack tip quickly
becomes comparable in size with the geometric residual stress field the effect
of residual stresses is likely to be small since the residual stress field will
be dispersed early in the life.
stresses. The results quoted in Appendix F suggest, in fact, that for crack
propagation tests in the absence of geometric residual stresses 1 ; may be
expected to be close to unity for Gaussian random loading in aluminium alloy
specimens but greater than unity for gust loading spectra where the ratio of
peak stress to rms stress is greater.
The method assumes that for notched and plain specimens the value of
The g5yst values of 1 ; for alum~nrum alloy were for direct stress
condltlons ' although nbt for the steel specimens. If this trend for alu-
mlnlum alloy specimens 15 slgniflcant, size would also be a factor since very
large specu~~ens in bending approach stress conditions over the area of action
of fatigue ldentlcal to that of specimens under direct stress.
"p 58
The effect of truncation level on 1 i has been investigated for
One other phenomenon which shows promise of being largely responsible for
test result inconsistency is the effect of residual stresses due to manufacture.
In section 3.2 it was shown how such stresses can be the cause of large varia-
tions in 1 + III. aluminium alloy specimens (see Fig.9) and can give rise to
values of l z greater than unity. It was also shown that mild and medium
strength steel specimens are generally unaffected in this way. Although only
56-58
three investigations on such steels are quoted in Table 1, and there are
differences in bandwidth and stress concentration factor, 1 $ values are
remarkably consistent. Therefore, while it is impossible to say with certainty
whether or not manufacturing stresses are responsible for the variation in
results of different investigations on aluminium alloy specimens, the evidence
quoted here is consistent with this hypothesis. If this is so, the results of
such investigations on aluminium alloys are of little use without any initial
measurement of the state of stress of specimens before testing. A number of
methods are available to do this, and a bibliography of papers on this subject
is given in Ref.59.
(Appendix D). The circumstances under which beneficial effects have been found
to predominate are stated in Appendix C. For such steel components too, maau
stresses do not greatly affect either absolute life or 1 i for the reasons
already stated 44 . Also stress concentrations do not greatly affect 1 g for
these materials. Hence it can be concluded that for non welded components in
low and medium strength steels present design practice is adequate in principle,
but it should be emphasised that it is advisable when using the Rule to assume
a value of 1% of about 0.5. This is put forward as au average obtained in
practice and not as a 'safe' value - the life so estimated will generally be
divided by a safety factor to cover scatter, corrosion, aad other effects. In
cases where a large proportion of stress cycles are below the fatigue limit a
lower value of 1 g should be assumed.
In section 3.1 it was shown that the local mean stress at the root of a
stress concentration in an aluminium alloy component under loading at a constant
amplitude o
a and a positive mean stress om is given by equations (2) and (3), '
21
(4)
lme ABC . The calculatiori of predicted life proceeds exactly as iii the pre-
"lous section except that curve ABC is used instead of curve DBC . If the
post yield O/E curve 1s fairly flad, as in the figure, there is little &ror
xwolved in assuming to be zero, since the strain is only of importance
ER
Insofar as It enables the stress at the maximum load in the spectrum and the
corresponding to be calculated. In Fig.10 the error in hem in making
C%SD
this assumption would be the difference between o and
1 S2 *
In cases where, due to a particular combination of residual stress and
mean stress, compressive yleldlng occurs at a lower amplitude than tensile
yielding, the procedure 1s the same as previously, except that the compresskve
stressjstraln curve must be used, and will uxrease with alternating
'Zm
stress amplitude from a negative value (cf. Flg.3).
of osm from that given by equations (2) and (3). As a general rule31 the
Effect becomes of significance approxunately when the local stress changes sign
after yielding in one dlrection, e.g. after tensile yielding some compredsitie
yielding wrll occur as soon as the stress becomes negative. Figs.42 - 44 show
predicted and achieved fatigue lives under random loading conditions in khe
programe described later in this Report. In this case the Bausthinger Effect
may be expected to be significant at stresses above approximately 2.5 ksi rms.
Its effect ~11 be to reduce the values of predicted lives above thid stress
level.
23
Cd) The predlction strictly only applies to the parts of the fatigue
life fully affected by residual stresses and so should tend to overestimate
their effect. Work 1s needed to determne the percentage of fatigue life so
affected.
In the foregomg sections It was assumed that the basic fatigue informa-
tlon used to calculate variable amplitude fatigue lives was in every case data
obtamed under constant amplitude condltmns, i.e. the S-N curve. Xethods of
life predlction have been devised for components subjected to symmetrical load
8,63,64 in which data obtained under a standard form of variable
spectra
amplitude loadmg were used m the place of the 5-N curve.
1
Ns = (5)
p1 p2 p3
N+iT+N+‘-’
1 2 3
(a) The basic data would be conditioned to some extent for the effects
of dlfferent relative damage rates, variation with stress of residual static
strength at failure and for cycles below the fatigue limit.
(b) The values of oemr for the basic data would be cIoser to those
exlstlng under the service loading action. Corrections for geometric residual
stress would then be smaller and as a consequence more accurate.
A 20 ton Avery Schenck resonant fatigue machIne was used (Flgs.13 and
Iii). This nachlne has been modified' to perform fatigue tests under either
stnt~onary randon, randam programmed or constant amplitude conditions.
26
reduce the required power output of the main drive amplifier and to avold
excltlng unwanted machine resonances. This last point was of particular
xnportance I* this xwestigatlon since the mechanical resonant system of the
. machlne in Flg.13 took the form of a beam, and so was a multi-degree of freedom
system. In order to increase the operating frequency of the machine, all tests
were carried out with the beam excited ln its first overtone node at 112 Hz,
rather than in its fundamental mode at about half this frequency.
For random programmed tests the programmer altered the root mean square
value of the Input signal to the machlne ln steps by switching the random slgna.1
through each of three attenuators ln repeated order.
Analysx of the random load pattern was accomplxhed by supplying the out-
put from the strain gauges to an Instrument which can count the number of times
each of ten Independently preset voltage levels 1s crossed in a posltlve-going
dIrectIon. This 1s a more accurate method of measuring the shape of the load
spectrum than the previously used8 method of monitoring the envelope of the
random loading waveforp. Typical measured load spectra are shown in F1gs.15
and 16.
As ln Ref.8, the load spectrum chosen for the random programmed tests "as
that ImplIed by the gust spectrum ln the Royal Aeronautical Society data sheets66.
For the purpose of this rnvestigation, a new synthesis was prepared (Fig.17)
to give a spectrum with a snwller relative number of counts at the lower stress
levels than that used previously. In this way the length of time to iallure was
reduced for the random programmed tests. Fig.17 also shows a comparison between
the synthesis used in this programme and that used previously. In Figs.16 and 17
o/h is a non-dlmenslonal amplitude parameter, where G 1s the actual stress and
h is that value of stress corresponding to a 5 ft/sec gust on the Royal
Aeronautical Society spectrum. The method of synthesis of the spectrum from
three random components is described in Ref.8. Table 19 gives details of the
programmed spectra used in the fatigue tests.
Chemical composrtlon and tensile test results for the DTD 5014 alum~~um
alloy are given II?. Table 2, and a typlcal measured stress/strain curve for the
materxtl 1s given in Flg.18. All the specimen material came from the same melt.
Specunen and pxn dunenslons are given in Flg.19. As can be seen, the
specimen is in the form of a double-ended lug. Failure in fatigue for such
specimens occurs at the points shown, where a stress concentration factor of
2.96 is close to an area of fretting. Sample specimens were tested to deter-
rune approximately the state of macroscopic residual stress at the point of
fatigue initiation. Strain gauges having a 2 mm gauge length were first
attached to the specimen as shown in Flg.l?. After the gauge had been connec-
ted to a strain gauge bridge, the bridge was balanced, a ~eweller's sawcut was
made, as shown, to relax any longitudinal resrdual stresses at the gauge to
zero, and the bridge rebalanced so as to measure the change in strain. This
residual strain value was used to calculate the longitudinal residual stress.
Fatigue tests were carried out by applying fluctuating tension at 112 Hz,
and all specimens were assembled without grease or antl-fretting compound
between the pins and the holes. For selected random and random programmed
tests the dlstributlon of levels crossed (spectrum shape) was measured. TypIcal
results of these analyses are shown in Figs.15 and 16*. These in general confirm
the measurements taken during the previous uwestlgation, demonstrating that at
the higher random stress levels the amplitude dlstributlon tended to truncate so
that few peak alternating stresses were obtained greater than o rovlded
m pp
Not untxpectedly, It was found that the stress spectrum truncated more
sharply on the compressive side than on the tensile side. Fig.20 shows the esti-
mated truncation levels at all the stress levels tested.
The results of the tests under the three types of loading are shown in
Tables 3 - 11 and Flgs.21 - 29. In all cases the stress value plotted is the
root mean square stress. Partial damage tests were conducted for constant
amplitude and stationary random loading by breaking statlcally specuwns which
had been fatigued to various percentages of the expected endurance. The falling
load was recorded in these tests and the change in residual strength from the
untracked state was regar-led as a measure of the fatigue damage. The results of
the PartId damage tests drf given =n Tables 12 - 17 and Figs.30 - 32 and were
*In the text, and diagrams, the mean and alternating rms stresses are shown, for
example as 10/3.5 which represents 10 ksi mean stress with 3.5 ksi rms alterna-
tmg stress.
29
used to construct S-N and o-N (stationary random) curves which applied to
different measured states of fatigue damage, i.e. 95, 90, 80, 70 and 60 per
cent of the untracked static strength (see Figs.21, 22, 24, 25, 27 and 28).
.
Partial damage tests were not carried out under programmed random loading. It
is hoped to do this at a later date.
Typical broken specunens are shown ln Flgs.39 and 40. It was found that
failure always occurred from an area of fretting close to the point of maximum
stress concentration.
n
The results show that ln this Investigation I, was generally greater
than unity under random and random programmed conditions (see Figs.23, 26, 29
and 33 - 35), ln line with a number of other investigations into cumulative
damage of aluminium alloy specunens with stress concentrations at a positive
mean stress (see SectIOn 3.4). Exceptions to this were found at the alternating .
stresses below 10/Z ksi random loading. Under random loading conditions at
10 ksi mean stress, 1 i was found to uxrease with rms stress at all states
of damage up to 3.5 ksi alternating stress, the highest value tested (Flg.36).
At 16 and 25 ksi mean stress 1 s increased up to a maximum value at about
2.5 ksi (Flgs.37 and 38) and thereafter decreased. Also under the stationary
random loading 1 ; values were generally greater for lnitlation than for the
total lrfe to failure, and were higher for random programmed tests than for
statzonary random loading.
The partial damage tests under constant amplitude loading also showed
that at all three mean stress levels the reduction in residual static strength
early in the life occurred relatively faster at low than at higher stress levels.
This 1s an effect opposite to that which has been found for plain speclmsns
(Appendu A)and there are three likely explanations for this. One explanation is
again the action of fretting. Fretting lowers the fatigue limit and so at
stresses below the unfretted fatigue lrmlt ~111 reduce the fatigue life from
virtually ufinlty to a finite value. Hence on a percentage life basis, fretting
1s more effective at reducing fatigue iritiation time at low than at high
stresses. Therefore fretting may reverse the trend of inltlal relative fatigue
damage rate with stress. Geometric residual stresses also provide an explana-
tion for the reversal in trend. In a notched specimen under constant amplitude
loading, at a positive mean stress, such residual stresses reduce the local mean
stress as the alternating stress is Increased. As a result of this, the rate of
fatigue damage initiation should not increase wzth stress in a notched specimen
as fast as in a plain specimen. Thjrdly, the fact that speclmans which are
fatigued at high stress levels will fail at the lowest state of damage
(Appendix B) , tends to cause the observed trend in damage accumulation. If, to
take an extreme case, a specimen were cycled such that the maximum stress in the
cycle was 99% of the ultimate, then very little fall in residual strength would
be detectable during the life. Hence at the lower stress levels the residual
strength against percentage life curve must show a greater drop, at least in the
later stages of the life.
32
There is some evidence from the partial damage tests to support fretting
as having some effect on the shape of the residual strength curves. From F1g.2
It can be seen that for Initially stress-free specimens at 10 ksi mean, under
constant amplitude loading, the local mean stress was unaffected by alternating
stresses up to 4 ksl rms. (Initial compressive residual stresses would increase
the unaffected stress range.) Fig.30, however, shows that, early III the life,
damage still accumulated faster at 2 ksl than at 4 ksi. Hence since residual
stresses were not effective, fretting was a lrkely cause of the particular
varlatlon of damage rate with stress measured here.
As for the effect of damage state at failure, Fig.41 shows the results
at 25 ksi mean, replotted to represent failure under all alternating stresses at
the state of damage corresponding to failure at 1.5 ksi rms. As can be seen, by
comparison with Fig.40, the general trends of the damage against percentage life
curves were not affected by the changed failure criterion, and would clearly also
not be affected at the other mean stress levels. Damage state at failure was
therefore not important in this case.
Under random loading, the residual strength against cycles curves show that
at 16 and 25 ksl mean, at the lower stress levels, damage did not accumulate 7
nearly as fast early in the life as it did under constant amplitude loading
at smnlar lives (see Figs.31 and 32). Different behaviour was shown with 10 ksi
mean stress where at 1.5 ksi rms the trend was similar to that found under con-
stant amplitude loading (Fig.30) and 1 z was less than unity. This was the
only random loading case III which the maximum static plus dynamic loads were not
high enough to cause yleldlng at the root of the stress concentration. This
dlfferent behaviour strongly suggests that the initial, relatively reduced,
rate of damage accumulation under random loading at the higher mean stresses
was associated with geometric residual stresses. It can be deduced that
residual stresses Increased the lnitlatlon time relatrve to the crack propaga-
tlon tune at the lower random stress levels at 16 and 25 ksi mean, to produce
the trend XI damage/percentage life shown. This 1s further demonstrated in
Flgs.36 - 38 in which 1 i IS shown to have been larger up to damage states
reached early III the life, than for flnal fallure except at lo/l,5 ksi, i.e.
1 i was greatest for the lnltlatlon.
n
33
In another respect the shapes of the o-N curves at 25 and 16 ksi mean
stress are consistent with the effects of geometric residual stresses (Figs.34
and 35). At the approxunate stress above which the Bauschinger Effect
.
(section 3.1) was operative the o-N curves stralghten or even bend slightly
towards the low life direction rather than maintain a concave shape as predIcted
by Miner's Rule. This produced a decrease in ; ; (Flgs.37 and 38). The pre-
dlctlons of fatigue life based on consideration of residual stresses as I*
section 4 show a sxmilar reflex shape for the predicted 0-N curves with the
point of inflectlon at a higher stress than that found in practice. Allowance
for the Bauschlnger Effect should lower the point of lnflectlon and so reduce
this discrepancy between the experImenta and predIcted results.
Finally, as in Ref.8, for the random programed tests at all mean stress
levels 1 ; to faliure was consistently greater than under stationary random
conditions. This general trend 1s also predlcted by the method of sectlon 4.
However, recent work has shown that 1' ; values greater than 1 can be obtaIned
69
for crack propagation under a gust load spectrum . It 1s therefore possible
that parts of the increases in 1 ; in the programmed random cases were in the
later stages of life.
The partial damage tests were used to obtain an estimate of the effect on
Miner's Rule of residual static strength at failure by means of the construction
of S-N curves representIng life to various values of static strength. Fig.42
shows S-N curves, at the three mean stress values, representlng life to a
static strength value given by the highest load in the random spectrum, at the
rms levels shown. Therefore, for example, If the life to failure at 25/6 ksl
random 1s required using Miner's Rule but allowlng for the effect of residual
static strength, then the Rule should be applied to the left-hand curve Instead
of that to fIna failure at 25 ksl mean. Table 18 shows the modlflcation due
to this phenomenon requred for the calculated lives. The largest effect can be
seen to have occurred at a mean stress of 25 ksl, at which the highest static
plus dynamic loads were applied. The conclusions reached in Appendix B (that
the effect on the achieved life of a bolted joint would only be large if loads
approaching the untracked static strength of the specimen were to be applied
during the fatigue life) clearly also applies in this case. One point however
needs clarlfxatlon. Although the effect has been expressed, in this case, as
a percentage change in life, in prlnclple the effect of this phenomenon is to
34
shorten the life by a given number of cycles rather than by a given percentage
smce the rate of fall off of residual strength 1s a function of crack growth
rate and does not depend on Initiation tlrlle. Therefore, If the total achieved
life 1s shortened by an unfavourable intcractlon in the initiatmn stage, the
effect of residual strength at failure ml1 be proportionally greater. In the
case of this mvestigatlon the Effect, as a percentage of the achieved life,
was generally less than shown 1~ Table 15, smce 1 s in most cases was
greater than one.
The results of r'ie tests carrled out under programmed random loading
(Flgs.26 and 29) lndlcate that at the higher mean Stresses the modzfied rule
(described =n Ref.8 and sectlon 4.6) applied to random data gave a signifi-
cantly better predlction than Miner's Rule based on constant amplitude data.
At 10 ksl mean stress (Flg.23), however, Mmer's Rule was marginally superior.
In all cases 1 i was greater than unity. The generally superior performance
of the modlfled rule can be explalned 3y the conditioning argument of
section 4.6. The questlon as to whether constant amplitude, preloaded con-
stant amplitude or SausSlan random loadmg data (see section 4.6) is the most
effzcient for use m life predIctIon can only be settled by further experi-
mental work.
Consider first the results of random tests at 10 ksl mean (Flg.33). The
interesting characteristic here is the near vertxal portIon of the Cl-h'
curve. As can be seen, the sharp upward trend in the curve was predIcted by
the rule assummg zero imtlal compressive residual stresses, but Its severity
was underestimated. Also, although the rule gave a good prediction for fma1
failure, this was probably comcldantal. In considermg this apparently good
prediction it should be noted that the plot of 1 $ against stage in life
(Fig.36) clearly indicates that the longer than expected life 1s due to the
fact that there is a very high value of 1 i early m the life and not a
constant value all the way through. This lnltlal value was underestimated by
the rule (see predxted results for 95% static strength - Flg.33), and the good
agreement for the final failure case was only obtained by assummg the full
residual stress effect, I.e. constant 1 $ over the whole life. When an
initial compressive residual stress was assumed of 10 ksl or greater, the
prediction became practically identical with that of Miner's Rule.
35
Under programmed random loading (Figs.23, 26 and 29) the new rule over-
estimated the life in every case for 11ves to final failure. Thu is to be
expected III "hew of the performance of the rule under stationary random loading,
and because the rule assumed Initiation condxtions throughout the life; as
dlscussed III section 3.3 residual stresses may be expected to be more effective
early III the life.
36
(1) Miner's Rule was unconservatlve under loading conditions which did
not cause local yielding.
(2) Where 1 z was greater than unity the unprovement was associated
with the earlier stages of the fatigue process.
(3) 1; Increased
. with increase in mean stress. The rule predicted
that the optimum effect is reached for initially stress-free specimens approxi-
mately when the mean stress alone just causes yxlding at the notch root. For
mean stresses above this value the magnitude of oLrn at a particular value of
alternating stress is almost independent of mean stress, and so little further
unprovement in 1 R should occur. The results of the investigation were
conslslstent with this.
6.6 scatter
so,70
In tests on plain specimens of aluminium alloy and high strength
stee171 It. has been found that scatter at the longer fatigue lives is signifi-
cantly less under random than under constant amplitude loading. In the present
mvestlgatlon, although not enough specimens were tested to obtain accurate
scatter factors It is clear that in general there was no significant reduction
under random loading. Scatter was generally low under constant amplitude load-
72
mg in these tests, as 1s usual for this type of specimen , where fretting
shortens the initiation time. This is to be expected because scatter is nor-
73
mally attributed to the crack initiation phase rather than crack propagation ,
an hypothesis which has been confirmed by crack growth measurements over the
68
entlre life of notched and unnotched aluminium alloy specimens . It therefore
follows that any improvement in scatter under random compared with constant
amplitude loading will be much less where fretting is present, since initiation
is a much smaller percentage of the fatigue life under fretting conditions.
37
It was noted earlier that the results show a variation in damage against
percentage life different in the random case from that found under constant
amplitude loading. This fact has practical implications in two directlons.
Cd) The random data may be used to predict the programmed random
11~s as suggested in sectlon 4.4.
(f) It is hoped that the work may also be extended later to obtain 'pre-
loaded constant amplztude data' so that this can be evaluated for use III life
prediction, as suggested in section 4.4.
8 CONCLUSIONS
8.2 An investigation into cumulative fatigue damage was carried out using
aluminium alloy lug specunens. The results of this work strongly indicated
that geometric residual stresses were the most significant cause of deviations
in 1 z from unity. Although a simple rule to account for these stresses did
not give good quantitative predictions, the qualltatlve agreement was good and
it is believed that a practical and reasonably accurate rule can be developed
along the same lines.
39
8.5 For aluminium alloy lug specimens the variation of residual static
strength throughout the fatigue life was found to depend greatly on loading
action and stress level. For this reason, when tests are carried out on an
aluminium alloy component to determine the variation of residual static strength
with life, a representative service loading action must be applied if misleading
results are to be avoided.
.
40
Appendix A
fatigue life consumed, (1.e. there are different shapes of damage against per-
centage life curves at different stresses) and:
then failure is predicted at 1 i < 1 , i.e. due to the above effect along Miner's
Rule should overestimate life.
The results of step tests on notched and plain specimens of steel and alu-
minium alloys (excluding alclad material) at zero mean stress indicate in most
Some life prediction methods have been based on accounting for the varla-
7,13,15
tion of relative damage rate wth stress and hence they predict
lives shorter than those predicted by Miner's Rule when applied to spectrum
loading.
alumimum alloy and steel specuwns, 1.e. in a step test the life at the second
level exceeded the life at that level for a “lrgln specxnen. The result of
this factor is to necessitate the drawing of damage curves wth ‘negatlve
damage ’ Ill scme Circumstances. Fig.45 shows the damage curves obtaned from
Flg.44, assuming a straight line relatlonshlp for level A , and demonstrates
the negative damage unpiled by the strenpthenlng effect. Fig.46 shows the cal-
culated results to be expected fron tw-level block programme tests on the
Sp~CUllfTlS, neglecting the strengthening effect by assumuxg the damage rate to be
zero when in fact It was negative. The oblect of this is to obtain a rough
estimate of the contrlbutlon to 1 i of variations xn damage rate alone. The
figure shows that the minxnun possible value of 1 $ was 0.67 between levels
A and D , a value which 1s unreallstlcally low due to the assumption made
about the strengthening effect. For spectrum tests with levels A and D the
predicted 1 g would zn any case be closer to unity. In this case then, for
spectrum tests the effect of dlfference III relative damage rate is not likely to
be large.
Appendix B
.
44
Appendix C
At first sight it would appear that cycles of stress in the spectrum which
lie below the initial fatigue limit should be damaging, since after a crack has
formed in a component, Its effective fatigue limit might be expected to drop and
the lower-level stress cycles become effective. Investlgatlons have been
80,91,95,96
carried out in which steel specimens were subJected to cyclic
pre-stressing above the fatigue lxoit. Subsequently the pre-stressed specunens
were used to obtain complete S-N curves. In all cases this treatment was
found to lower the effective fatigue limit which confirms that the above sltua-
t1on can apply In
some circumstances. The same behaviour has also been found to
80
apply to an alumlnlum alloy . In Refs.11, 15 and 80 cumulative damage rules
are described which account for this behaviour.
91 that
It has, however, been shown for steels cyclic pre-stressing at a
single stress level below the initial fatigue lunit can actually raise the effec-
tive fatigue llnnt. In addition, when subJected to a stress hlstory in which the
stress started at a low alternating stress level and was raised in increments to
a value above the fatigue lxnlt, steel specimens have at that raised level sur-
vlved many txnes the endurance of virgin specunens. This second phenomenon is
76,78
known as coaxing. Alumln~um alloy specimens do not exhibit this property
97,98
However It has been shown that periodic stressing below the fatigue limit
(1.e. two-level block programme test with one stress below the fatrgue limit)
can increase the life at the higher level for aluminium alloy specunens. 1 i
values greater than one have been obtalned also in programmed tests on steel
99 .
specimens with large numbers of cycles below the fatigue limit
100,101
Coaxing in steel specimens has been attributed to strain ageing
Alumlnlum alloys, which are not strain ageing materials, do not show coaxzng
behavuwr. It has, however, been shown 7,75,76,78,79 that both steel and
alununlum alloy specunens show sunllar strengthenlng behaviour in low-hlgh
step tests, where the low stress 1s above the lnltlal fatigue limit. Thu
suggests that there exists a mechanism in addition to coaxing, by means of whxh
damage at high stress levels 1s suppressed by cycles of low amplitude. This is
unlikely to involve macroscopic residual stresses as described in section 3,
since the action of these 1s to change 1 i by the effect of high on low stress
cycles.
Appendix C 45
.
46
Appendm D
Low-level stress cycling (i.e. generally below the yield point) has also
been shown to give rise to both strain hardening and strain softening. In one
104 a range of metals
investigation in the annealed condition were shown to
exhlblt cyclic strain hardening propertIes under such loading, whereas in the
cold drawn condition the opposite was true. As for the connnon engineering
materials, mild steel shows marked cyclic strain softening behaviour and has
been found to develop considerable plastic flow at alternating stresses of 75%
of the initial yield point corresponding to failure in about lo6 cycles. For
26s aluminium alloy the same effect was found at stresses for failure in the
41
same number of cycles, but the effect was very much smaller .
.
48
Appendix E
There are also other possible effects of fretting which act in opposition
and are both also connected with residual stresses. Since the effect of such
Appendix E 49
.
50
Appendix F
The fatigue process m plam specmens has been classified into two
110
stages . Stage I starts from the surface of the materlals and 1s dependent
upon the maxumxn resolved shear stress. Crack growth occurs along the plane of
0
maximum shear stress, at 45 to the maximum tensile stress. The stress state
at the tip of the crack changes as the crack begms to propagate from a bm.xiaI
to a trlaxlal state of stress. The effect of this is to reduce the ratio
maximum shear stress
and to change the fatigue mechanzsm to Stage II
maxlm”m tens11e stres?.
behawour. Crack growth m thu second mode occurs, at least initially, at
rlgnt angles to the dIrectIon of maxmum tensile stress, and is characterised by
striation markings at right angles to the dlrectmn of growth and a l/l rela-
tmnshlp between number of strlatlons and number of cycles. All the investiga-
tlons referred to m thu section apply to Stage II crack propagation in
fluctuating tensLo*.
Two-level constant amplitude tests have been carried out on cracked panel
111
specimens of 2024-T3 alclad materml . It was found that when the stress was
changed from a high to a low level, crack propagation was retarded for some time, .
gradually acceleratx,g to the value which would have been expected from a con-
stant amplitude test at the lower level alone. in opposite though smaller
.
effect was found when changing from a low to a high level, 1.e. after the cnange
the crack propagation rate accelerated to a rate somewhat beyona the cal.ie tz be
expected from the higher level before settling down to the value expected frcn
constant amplitude tests. It was also found that single tensile loads ln the
middle of a constant amplitude loading pattern caused a temporary retardatwn.
Single negative loads did not accelerate crack growth, but If applied after a
single tensile load they reduced the retardation due to that load. The retar-
datlon of the crack growth rate m the high-low sequence was attributed to a
region of residual compressive stress Induced at the crack tip by the high
Initial loading. It was also assumed that the actIon of any succeeding negative
load was to destroy the residual stresses in the region by local yielding. The
overall effect ga”e values of 1 ; greater than one for crack propagation rates
under programmed and subsequently (Ref.69) under random loading to a gust load
spectrum.
112
Under Gaussian random loading however lt has been found that under
direct stress conditions in certain cases crack propagatun rates can be
Appendm F 51
.
52
Appendix G
(7)
where D = damage
n
= cycle ratm - percentage of life consumed at the lth level
N1
-=dD
dn
1
x1-1
dD
-=pqj;
drib
x1
0 "1
.
Appendix G 53
1
dD
Nb =
I0 1-2
+- p2 22,
( x1 >
N2
therefore
dD
Let
p1
q
-+-
=
.
[ 1
Pl
N1
p2
N2
therefore
dD
C X1 D + (1 - C) X2 D
Let
xl
D = CD')
.
54 Appendix G
therefore
.
(x1-1)
1; = Xl@‘) dD’
f
0
(X1-l)
C3 D' + (1 - C) X2 D'
0
J dD'
(8)
x2
C+(l-C)F
1
Therefore 1 i x2
depends only on ii- for two-level programmed tests.
1
Equation (8) has been evaluated by computer programme. The results are shown
in Fig.43.
. x1
“1q
Remaining life at level 2 = N
2 L- -N1
therefore
!)I
achieved life =
life predicted from Miner's Rule (9)
and
00)
*
achieved life
Note that for step tests the ratio is not equal to
predicted life
1; except when 1 z = 1 . This is because if Miner's Rule does not hold,
the component fails under a different load spectrum (i.e. different proportions
of counts at levels 1 and 2) from that which it would fail under if Miner's
Rule did hold.
56
Table 1
Table 2
__-
XEASUREDCHEMICAL COMPOSITIOL~
AND TEXSILE PROPERTIESOF
DTD 5014 AL~IIIIIW ALLOY
Balance aluminium
Average tens11e properties
UTS 58.5 ksl
Table 3
STATIONARYRANDOMFATIGUE TEST RESULTS
MEAN STRESS = 10 ksl
Stress ksl rms Spec. No. Life (N)
1.0 1213-F78R 1.79 lo7
1.0 0408-F84R 1.35 lo7
1.0 1214-F85R 4.00 IO7
1.0 0409-F86R 2.20 lo7
1.0 1806-F367R 6.92 LO6
1.0 1922-F370R 7.90 lo6
1.0 2119-F371R 8.20 lo6
1.0 2007-F372R 9.50 lo6
1.0 0119-F399R 7.27 LO6
1.0 0222-F408R 9.01 lo6
1.5 0404-F76R 3.70 lo6
1.5 1216-F82R 3.97 lo6
1.5 2121-F35CR 2.70 lo6
1.5 1711-F373R 3.20 lo6
1.5 2403-F374R 2.70 lo6
1.5 2314-F437R 2.29 lo6
1.5 2811-F439R 3.66 lo6
1.5 3616-F441R 2.75 lo6
1.5 3221-F443R 1.78 LO6
2.0 0411-F74R 1.26 lo6
2.0 1203-FBOR 1.08 LO6
2.0 2407-F349R 9.80 lo5
2.0 2318-F366R 2.11 lo6
2.5 0414-F72R 9.00 lo5
2.5 1217-F81R 8.72 lo5
2.5 2321-F348R 9.30 lo5
3.0 0419-F73R 8.66 lo5
3.0 1220-F77R 6.56 lo5
3.0 041%F142R 8.96 lo5
3.5 0403-F75R 8.00 lo5
3.5 1202-F79R 8.40 lo5
3.5 1215-F83R 7.56 lo5
3.5 0410-F87R 8.30 LO5
3.5 1619-F345R 8.38 LO5 ,
3.5 2215-F365R 6.91 lo5
59
Table 4
Table 5
RANDOMPROGRAMMED
LOADING FATIGUE TEST RESULTS
MEAN
.,- .- STRESS = 10 ksl
Stress ksi rms Spec. No. Life (N)
1.230 0805-F180RP 1.26 lo7
1.230 1211-F181W 8.36 lo6
1.230 1602-F182W 1.68 107
1.230 1003-F183W 1.05 107
1.230 0914-F184W 8.11 lo6
Table 7
stress
Spec. No. Life (N) Spa. No. Life (N)
ksl rms
Table 8
RANDOMPROGRAMMED
LOADING FATIGUE TEST RESULTS
MEAN STRESS = 16 ksi
.
64
Table 9
STATIONARY RANDOMFATIGUE TEST RESULTS
MEAN STRESS = 25 ksi
.
65
Table 10
Table 11
RANDOMPROGRAMMED
LOADING FATIGUE TEST RESULTS
MEAN STRESS = 25 ksl
Table 12
STATIONARY RANDOMPARTIAL DAMAGETEST RESULTS
.
68
Table 13
Table 14
Seconds Fallmg
S?ax. No. to % stopoff load
stopoff (ton)
.
70
Table 15
.
71
Table 16
Table 17
f
73
Table 18
Predicted 1 i values, allowing for failure at the highest load in the spectrum
Random loading
10 2.5 0.95
10 3.5 0.95
16 2.5 0.96
16 3.5 0.95
16 5.0 0.95
25 1.5 0.95
25 3.5 0.90
25 5.0 0.87
Table 19
PARAMETERSFOR PROGRAMMED
RANDOMLOADING
0 996 1.197 1.395 1.595 1.795 1.984 2.19 2.39 2.59 2.79
REFERENCES
REFERENCES (Contd.)
.
76
REFERENCES (Contd.)
REFERENCES (Contd.)
REFERENCES(Contd.)
REFERENCES(Contd.)
REFERENCES (Contd.)
REFERENCES- (Contd.)
.
& Author Title, etc.
lretals.
Department of Mines and Technical Surveys, Ottawa
theories.
Rand Corporatmn. Memorandum FX-3650-PR.
August 1963
hole.
tenyieratures.
REFERENCES (Contd.)
REFERENCES (Contd.)
REFERENCES (Contd.)
REFERENCES(Contd.)
30
I I I
20
IO
0
0 2 4 6 8 IO I2 14
rms net stress (Ksi) c
Fig.2 Calculated variation in d’m for small lug specimens under constant
amplitude loading neglecting Bauschinger effect
” . P
I I I I I 0
I
a.
.
*
I
I6
I
s
0---
.-CT 6-
4
I”
Number
106
of cycles
.-
IO7
- NA
108
R
ul X’
I2 -2 $
.- a
2 9
8 Qa 0
0
\
\
4
0
7
la15 I06 10s
Number of cyc\er - N -
50
I
X
K\
1;
.- 40 \
~ \ X
2
I
\
-X-
30 k-z
-o-
20
I
X Stress -free
-0 With residual streis-
IO
Number of cycles - N -
?OOO - 10000 00 -6
,I0 -
20 -
30 -
40 -
,50 ___
Fig.7 Behaviour of annealed mild steel
a Crack growing from notch In an infinite sheet
1 -t t
- Length at
d
fatlure
e for constant
I etude loading at
I I I I IIII
I I I III1
Cycles to failure *
Fig.9 Set of fictitious S-N curves
W
L
-----
-----
__---e
-----
----
---
.-- L
--
_---
---
--
__L--_
-
D
.
I
I
I
I
I
1
I
I
I
L&ga---
2L
__------
_------
*0 __------
__-----
3
_---
.----
___---
r- I
__------ Y
*
I
I
I
I
I
r--s _----L----
I
I
I
I
I
I
i
l
Time
07
36
05
0.4
03
02
l3cmd pass
“Z
voltmeter
r-b Resonant
mochme
23
9
Controller
5
2
s
proqrommer
- ‘c-------L
-01
-001~
---
. 0001
50
40
20
IO
0
4000 8000 12000 16000 20 000 24 000 28000 32000
Strain Pe +
-5004
C- SOOIdia.
jeweller’s Sawcut
\ +&e length
2- Holes 1 E”o% dia ream of strain gauge
sym. tol. - ooz datum ‘A’ 4oor residual
stress measurement
Fig.19 Lug specimen and pin
T
I 7-
//
/
0
2 4 6 6 IO 12
matAg
I4
I
stre,ss
16
1
tcqi
18
I
20 22
I*
24 26 28
1
30
5 - -
I
Alternating
stress q
4 -
3ercentages of the
..
r
2 - -
I -
B!
90 !dd%%6
t
0 - I
Cycles
I
II5
N-
. li _
I
1
'u
a
s - N curves to diCCerent
percentuqes 04 the origina
static tenslle strenqth .
.stresS CT
(ksi rms) 1
IIT -
+ First
econd
test ‘series
test server 5
I. . u .
. c TV
/,’
).’ /
/
f
r //’
0,
r .o
I I
1 i
I ,
Li-’
- - - II,
I I
I I
- h
All akernatln stresses are rms
60
t
Net Static Perc=:fto=es
fabling 50 untracked
stress -Foi I;ng
(ksi) StreSS
40
0
0 of4P;Ce <Y*llw,““- 100
N.;bl~tL~ic So
. .
stress
(Ksi) 4o
IO
--- f 3.5 KSI (8.0 lo5 )
--@ I.5 KS1 (2 9 IO”)
01 I I I I I
0
70 04YXe t5a 1lur:LZ
60
Perc$ageS
uncrac ked
50 Calling
T! 4c
stress
Net
faillnq
stress 3o
(ksl)
20
10
C 80 100
o/o Zf Ilfe4L f411Ze -
Constant amplitude loading
60 I-
35%
-. 30 “/o
50 I-
8070
70”/0
t 4c I-
Nkt 600/o
faillnq
stress 3~ ,- --t-t-t-l--
(kS1)
ec I--
C
‘0 IO 0
Net
qai Iing
stress
(ksi )
O
Constant amplitude loading
100
O ,/aZDIi~~ 2 -Failu6roe 84
Stationary random loading
Fiq. 32 Partial damaqe tests 25 ksi mean
small DTD 5014 luqs
co
o_
i
w
.-E
4
0
\\ \ /I Ii
s \
2 \ 1
Cycles NW
Fig.34 Performance of cumulative damage rules for stationary
random loading at 16 ks~ mean stress
. u
x
S
3
i
Alternahg
stress Qt
(ksi rmS)
90
-A.-.- 80
-a.--. - 70
Fig 39. Specimens broken statically after fatigue to stated percentage of nominal life
‘8
50
40
\
\
t
‘\5
Net
f a1 lurQ
stress
Ksi 30.
‘ Mean stress = 25 Ksi
constant ampI1 tude
loadin ’
l I3
20
IO
0
0 2 40 60 8 1 I
5
Akernating
stress 0;
(ksi rms) 35
,si
4 -
l5-
Ksi
r\l cycles 8
I I1111.. IO6
I - - Programmed tests
16
-.-- Step tests I
I -----step tests I
15
level I D= Under consta
14
zx
13
12
I I
IO
c
9
e
-6
-5
I
_ ‘
effect to left
06
kef 75
I
Mat1 SAE 434.0 Steel
Level Stress
(ksi Peak) (k$es)
04 A 127 9 147 50
B 117.3 37 100
C 106 6 83800
D 95.9 234500
.R
U Percentage life -
o-9
o-8
0.7
tf
This graph has a Take zero
f
06
0
I
0.1 0.2 0.3
I 0 4
I
05 06 07
I 08
I
0.9 I.0
t
Damage at -First level according
to Mbners rule -
.
Ref 94
-- -- \
\
mean load ,!
I
--a
4%
-
)I
t- se
60 E3C )O
Fraction of \lfe y.
tY88’129
: SIL-699
: 61E’6ES
: IEY6ES
,
L------o --a------ A!
---------------
Pubhshed by
HER MAJLSTY’S STATIONERY OFFICE